Rendering times - help me to understand please

[kongc]

Looking for professional help here for my understanding. Does my conclusions make sense? I am trying to understand hardware encoding, vs cuda vs CPU software.

So this discussion / question would be would the pros agree or disagree with my results. I am coming from a bang for buck perspective. This is rendering H264 only

Don't need to consider storage as using 2x SSDs in both systems so I don't think storage would be bottle necking

Not sure if I need to consider ram as the 16gb ram is nearing but not topping out.

I am not doing per second timed benchmarks as real world results matter to me more.

I am rendering say 3 minute 4K clip with Lumetric effects - lets say 90% 2D stuff (hence I don't think cuda is used that much ??)

Rig 1 - i7-4770 @ 3.4ghz (8 threads) - 16gb ram - GPU = built-in Intel 4600 + Titan X (maxwell)

Rig 2 - Xeon 20 core server @ 2.4ghz (40 threads) - 256gb ram - Titan X (maxwell)

What I found is it is all about intel quicksync

Fastest I can do 3min 4k clip

i7 -4770 with quicksync using NEVEC codec - titan X maxwell   (about 5 minutes)  using cuda

i7-4770 with quicksync using built-in H264 - titan x maxwell (about 9 mins)

i7-4770 with quicksync using built-in H264 - gtx 680 (about 9 mins)

Xeon server would render using - titan X ( about 25 mins) using cuda

I am drawing the conclusion that for H264 using NEVEC codec the best bang for buck is to use i7 processor with intel quicksync and nvidia card

I was surprised that even putting in a Titan X on Xeon server with 20 cores and 256gb ram would be about 5x slower than i7-4770 with quicksync with H264

What do u guys say? Since quicksync is free with almost any currentish i7 cpu

[RjL190365]

Here's the reason why:

Large portions of the Premiere Pro program code are still only slightly threaded. As a result, Premiere Pro cannot effectively utilize more than about 10 cores / 20 threads total. Beyond that, Premiere Pro performance relies almost entirely on the actual CPU clock speed.

[kongc]

I do see the 20 cores 40 thread Xeon all threads used equally @ ~ 80% ... does this still mean only 10 cores being used?